I would suggest tweaking the test output to include regress_du_admin ...
I found (with a help of cfbot) difficulty with this. The problem is the bootstrap superuser name (oid=10).
This name depends on the OS username. In my case it's pal, but in most cases it's postgres or something else.
And the output of \du regress_du_admin can't be predicted:
\du regress_du_admin List of roles Role name | Attributes | Member of
------------------+-------------+------------------------------------- regress_du_admin | Create role | regress_du_role0 from pal (a, i, s)+ | | regress_du_role1 from pal (a, i, s)+ | | regress_du_role2 from pal (a, i, s)
So, I decided not to include regress_du_admin in the test output.
Please, see version 5 attached. Only tests changed.
-----
Pavel Luzanov